How to handle fields enclosed within quotes(CSV) in importing data from S3 into DynamoDB using EMR/Hive

问题

I am trying to use EMR/Hive to import data from S3 into DynamoDB. My CSV file has fields which are enclosed within double quotes and separated by comma. While creating external table in hive, I am able to specify delimiter as comma but how do I specify that fields are enclosed within quotes?

If I don’t specify, I see that values in DynamoDB are populated within two double quotes ““value”” which seems to be wrong.

I am using following command to create external table. Is there a way to specify that fields are enclosed within double quotes?

CREATE EXTERNAL TABLE emrS3_import_1(col1 string, col2 string, col3 string, col4 string)  ROW FORMAT DELIMITED FIELDS TERMINATED BY '","' LOCATION 's3://emrTest/folder';

Any suggestions would be appreciated. Thanks Jitendra

回答1:

If you're stuck with the CSV file format, you'll have to use a custom SerDe; and here's some work based on the opencsv libarary.

But, if you can modify the source files, you can either select a new delimiter so that the quoted fields aren't necessary (good luck), or rewrite to escape any embedded commas with a single escape character, e.g. '\', which can be specified within the ROW FORMAT with ESCAPED BY:

CREATE EXTERNAL TABLE emrS3_import_1(col1 string, col2 string, col3 string, col4 string)  ROW FORMAT DELIMITED FIELDS TERMINATED BY ',' ESCAPED BY '\\' LOCATION 's3://emrTest/folder';

回答2:

I was also stuck with the same issue as my fields are enclosed with double quotes and separated by semicolon(;). My table name is employee1.

So I have searched with links and I have found perfect solution for this.

We have to use serde for this. Please download serde jar using this link : https://github.com/downloads/IllyaYalovyy/csv-serde/csv-serde-0.9.1.jar

then follow below steps using hive prompt :

add jar path/to/csv-serde.jar;

create table employee1(id string, name string, addr string)
row format serde 'com.bizo.hive.serde.csv.CSVSerde'
with serdeproperties(
"separatorChar" = "\;",
"quoteChar" = "\"")
stored as textfile
;

and then load data from your given path using below query:

load data local inpath 'path/xyz.csv' into table employee1;

and then run :

select * from employee1;

Now you will see the magic. Thanks.

回答3:

Following code solved same type of problem

CREATE TABLE TableRowCSV2(    
    CODE STRING,        
    PRODUCTCODE STRING, 
    PRICE STRING     
)
    COMMENT 'row data csv'    
ROW FORMAT SERDE 'org.apache.hadoop.hive.serde2.OpenCSVSerde'

WITH SERDEPROPERTIES (
   "separatorChar" = "\,",
   "quoteChar"     = "\""
)
STORED AS TEXTFILE
tblproperties("skip.header.line.count"="1");

回答4:

Hive now includes an OpenCSVSerde which will properly parse those quoted fields without adding additional jars or error prone and slow regex.

ROW FORMAT SERDE 'org.apache.hadoop.hive.serde2.OpenCSVSerde'

回答5:

Hive doesn't support quoted strings right out of the box. There are two approaches to solving this:

Use a different field separator (e.g. a pipe).
Write a custom InputFormat based on OpenCSV.

The faster (and arguably more sane) approach is to modify your initial the export process to use a different delimiter so you can avoid quoted strings. This way you can tell Hive to use an external table with a tab or pipe delimiter:

CREATE TABLE foo (
  col1 INT,
  col2 STRING
) ROW FORMAT DELIMITED FIELDS TERMINATED BY '|';

回答6:

Use the csv-serde-0.9.1.jar file in your hive query, see http://illyayalovyy.github.io/csv-serde/

add jar /path/to/jar_file

Create external table emrS3_import_1(col1 string, col2 string, col3 string, col4 string) row format serde 'com.bizo.hive.serde.csv.CSVSerde'
with serdeproperties
(
  "separatorChar" = "\;",
  "quoteChar" = "\"
) stored as textfile
tblproperties("skip.header.line.count"="1") ---to skip if have any header file
LOCATION 's3://emrTest/folder';